Model-Based SVM Parameter Selection Approach for Multi-View Fall Detection

Falls indeed contribute significantly to injury-related fatalities among the elderly population worldwide. Thus detecting falls efficiently is crucial for reducing risks of harm. This process of detection is in general achieved through wearable gadgets or environmental sensors which encounter challenges such as user adherence or false alerts. Alternatively, video cameras offer an interesting passive solution, yet susceptible to lighting changes and privacy issues. From a machine learning standpoint, crafting a reliable fall detection system poses difficulties due to the infrequency and diverse nature of falls. Among the important challenges and issues reported in literature is the difficulty of fair comparison and analysis between fall detection machine learning techniques. This challenge arises due to variations in datasets, evaluation metrics, and experimental setups across studies. This paper utilizes a model-based approach to select hyper-parameters for training Support Vector Machine (SVM) models in the context of failure detection. Experimental results demonstrate that this method enables the discovery of parameter sets that yield higher values for two evaluation metrics: accuracy and F1-score. The proposed model-based approach effectively selects hyper-parameters for SVM models in failure detection, resulting in higher accuracy and F1 scores. It outperforms existing solutions and algorithms in predicting fall detection.